https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113682
Tamar Christina <tnfchris at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |tnfchris at gcc dot gnu.org
--- Comment #4 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
note that on aarch64 the situation isn't that simple and whether the csel or
branch is better in the loop depends on a number of factors.
We did extensive benchmarking on this on various uArches and there are a number
of cases where the csel inside the loop ends up being significantly worse and
the branch misprediction penalty also highly depends on the uArch.
In the end the conclusion was though that to have chance to do this properly we
would have to form conditional operations at the GIMPLE level and not in RTL as
we do today.
